def test_log_factor_hessian(self):
delta = 0.00001
y1 = 1.0
y0 = 2.0
theta = -2.0
alpha = 1.0
beta = 0.0
c = 0.25
# simulate the gradient
true_hessian_approx_theta = (tools.log_likelihood_factor_gradient(y1, y0, theta + delta, alpha, beta, c) -
tools.log_likelihood_factor_gradient(y1, y0, theta, alpha, beta, c)) / delta
# calculate
calc_hessian = tools.log_likelihood_factor_hessian(y1, y0, theta, alpha, beta, c)
self.assertTrue(abs(calc_hessian - true_hessian_approx_theta) < 1e-4)
def test_log_factor_gradient(self):
delta = 0.00001
y1 = 1.0
y0 = 2.0
theta = -2.0
alpha = 1.0
beta = 0.0
# simulate the gradient
true_gradient_approx_theta = (clib.log_likelihood_2PL(y1, y0, theta + delta, alpha, beta) -
clib.log_likelihood_2PL(y1, y0, theta, alpha, beta)) / delta
# calculate
calc_gradient = tools.log_likelihood_factor_gradient(y1, y0, theta, alpha, beta)
self.assertTrue(abs(calc_gradient - true_gradient_approx_theta) < 1e-4)
# simulate the gradient
c = 0.25
true_gradient_approx_theta = (clib.log_likelihood_2PL(y1, y0, theta + delta, alpha, beta, c) -
clib.log_likelihood_2PL(y1, y0, theta, alpha, beta, c)) / delta
# calculate
calc_gradient = tools.log_likelihood_factor_gradient(y1, y0, theta, alpha, beta, c)
self.assertTrue(abs(calc_gradient - true_gradient_approx_theta) < 1e-4)